International

OPEN ACCESS J ournal

Of Modern Engineering Research (IJ MER)

| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 34 |
Palmprint Identification Based on Principle Line Using Machine
Learning Techniques

Krishnaveni.M
1
, Arunpriya.C
2

1,2
(Research Scholar, Assistant Professor, Department of Computer Science, PSGR Krishnammal College for
Women, India)












I. Introduction
The security of a system has three main components of authentication, authorization, and
accountability. Authentication is the most important of these three elements. In the information technology
domain, authentication means either the process of verifying the identities of communicating equipment, or
verifying the identities of the equipments users, which are primarily humans. Security plays a very important
role in ones life. The accurate identification of the person to access secured application is static challenging due
to the limitations imposed by real time applications. Samples of such applications include access to ATM,
nuclear facilities, boarding a commercial flight or performing a remote financial transactions etc. The main goal
of accurate identification is to prevent the imposter accessing the safe application. There are three methods in
which users can be identified such as:
1. Something the user knowsPassword, PIN
2. Something the user hasKey, Cards and Tokens
3. Something the user isUnique Biological properties

Easily lost, stolen, shared or manipulated and there by undermining the intended security. The third
way of identifying the person appears to be more secure, so designing a security system based on biological
properties cannot be lost, manipulated or stolen. Biometric system can be defined as a pattern recognition
scheme, which is capable of finding a Person based on their biological properties. These biological properties
can be physical characteristics like face, finger print, hand geometry, palm print, retina, and DNA, ear, iris and
hand vein .Behavioral properties like speech, gait and signature.
A biometric is a physicalor behavioural feature or attribute that can be measured. It can be used as a
means of proving that you are who you claim to be, its means of proving without revealing your identity that
you have a definite right.
Apalm print refers to an image acquired of the palm region of the hand. It can be either an online image
(i.e. taken by a scanner, or CCD) or offline image where the image is taken with ink and paper. The palm itself
consists of principal lines, wrinkles and ridges. It differs to a fingerprint in that it also contains other information
such as texture, hollows and marks which can be used when comparing one palm to another.
Palm has unique distinguishing line patterns which can be used to identify people uniquely. It is a
physiological biometric. A user is asked to put his/her hand in an surrounded unit having fixed illumination and
a platform that restricts hand motion. It may have pegs to further fix the position of the hand. Scanners or
cameras can then be used to capture the image. Palm print has high user acceptability with few hygiene based
concerns characteristic of all contact based biometric. But there has been a move towards developing contactless
recognition systems.

Abstract: In this paper, we propose principle line based Palmprint Identification method. In this method
to detect principle lines of palm print is with consecutive filtering operations. Smoothing operation is used
to remove image noise. Edge detector operation and closing operation are merged to extract the principle
lines. Binarization yields the binary principle line. The lines detected with the developed scheme are used
to extract textural information using Gray Level Co-occurrence Matrix and Statistical Property Features.
Euclidean distance is used for matching to identify the genuine person and the powerful supervised
classification techniques namely Support Vector classification Machine and Extreme Learning Machine
with kernels like linear, radial basis function is applied to classification. The experimental results on the
PolyU palmprint database demonstrate the feasibility and effectiveness of the higher accuracy and
reduced execution speed shows that our proposed approach.
Keywords: Biometrics, palmprint extraction, palmprint identification, Principle lines, Machine learning.
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 35 |
There are two types of palm print recognition research,
1. high resolution
2. low resolution

The high resolution approach employs high resolution images, and is suitable for forensic applications,
such as criminal detection. In general, high resolution refers to 400 dpi [1].In high resolution images,
researchers can extract ridges, singular points and minutia points as features can be detected. The low resolution
approach employs low resolution images, and is more suitable for civil and commercial applications such as
access control. Palm print images can be acquired with low resolution cameras and scanners and still have
enough information to achieve good recognition rates. Low resolution refers to 150 dpi or less. The Low
resolution images are generally use principle lines, wrinkles and texture.


Fig.1 Palm print features in (a) a high resolution image and (b) a low resolution image.

Recognition by iris or retina brings very good results but tolerate very expensive devices of images
capture in one hand and on the other hand the possibility of intrusions among the users .Recently, many
researchers used verification by face or voice, unfortunately their performance is again far from being
satisfactory.
Another technology is based on the geometry of the hand. It uses the geometric information to identify
a person, unfortunately these informance are very limited with low accurate results. To face the problems
encountered in the biometry based on the hand, David Zhang and shuproposed, in 1996, another biometric
always based on the hand that is palm prints.
The palmprint, this very large and internal surface of the hand, contains several characteristic features
as principle lines, wrinkles, creases and textures. Thanks to this large surface and the abundance of characteristic
features, it seems that palm prints are very robust to noise and unique to every individual. Compare to other
physical characteristics.
Palmprints identification has several advantages
1. Treatments of low resolution images
2. Few intrusion risks
3. The features of lines are stable
4. High level of acceptance by users.

The Physical characteristics can be fingerprint, palmprint [2,3], face, iris, signature, voice, and gait.
Palmprint is an efficient characteristic because of its uniqueness and stableness. In addition, this characteristic
gives a lot of biometric information, e.g., principle lines, wrinkles, ridges and datum points which are the unique
features of human.
Palmprint-based biometric technology is composed of two steps: feature extraction and recognition.
Image features are extracted and collected as input vector while the remaining step is to recognize or classify
that vector into the suitable class to identify people.
In recent survey, Sakdanupab et al. [4] have proposed and implemented a palmprint classification
method based on principle lines. The phase of principle line extraction is based on profiles of gray values within
a window of size 3x3 in four directions (0, 45, 90, and 135 degrees). The principle lines consisting of heart line,
head line and life line are extracted and used for recognizing people afterwards. The drawback of this method is
that noise is not completely eliminated by the proposed noise reduction process. In addition, their algorithms
take too much processing time. The accuracy was achieved 85.49%.
Huang et al. [5] have proposed a palmprint verification approach based on principle lines. In the
process of principle line extraction, the lines are extracted by using the modified finite Radon transform. When
the transformation is applied, lines in Cartesian coordinate are converted to lines in energy and direction. The
energies and directions are used to detect the differences between principle lines and wrinkles. After that, those
differences are finally used to verify people. The accuracy was achieved 95%.
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 36 |
Wong et al. [6] have described the palmprint identification based on features from lines in palmprint. In
this work, Sobel mask of two sizes (3x3 and 5x5) in four directions together with an appropriate threshold are
mainly used to extract feature vectors. Proceedings of the Second International Conference on Knowledge and
Smart Technologies 2010 (July, 24-25, 2010)
In 2009, Zhu et al. [7] have applied an approach for major line extraction of palmprint using gradient
images in 4 directions which are overlapped, and then merged with canny edge image. To obtain better results,
palmprint image will be dilated and blurred with the probable template to get the major lines. The disadvantage
of this method is that their algorithm is somewhat complicated to generate the principle palmprints. From our
recent survey mentioned, the methods belonging to Huang and Wong are not used the principle lines as main
features. The methods attempt to extract features from the whole image instead. The accuracy was achieved
87.04%.
However, the methods from Sakdanupab and Zhu focus on extracting the principle lines directly. This
causes the problem because of their complexity and high processing time.In this work, the compact method to
detect principle lines of palmprint is proposed with consecutive filtering operations. Smoothing operation is
used to remove image noise. Edge detector operation and closing operation are merged to extract the principle
lines. Binarization yields the binary principle line.Finally, the postprocessing, removing the connected
components with small region, is applied to discard noise.
The Leqing Zhu and RuiXing, proposed a new hierarchical palmprint recognition method [8]. First the
gradient images along the four directions are computed. Then these four gradient images are overlapped and de-
noised. Edges are detected with canny detector and merged with the de-noised gradient image with and
operation. The result is then dilated and blurred with a probable template to get the major line features. The
bidirectional method is used for matching.
The Wei Jia, Yi-Hai Zhu, Ling-Feng Liu and De-Shuang Huang, proposed the palmprint retrieval
based on principal lines for palmprint recognition [9]. In this principal lines are extracted using modified finite
radon transform method. Then key points of principal lines are detected. And direction, position and energy of
these are stored in the table. During matching palmprint is retrieved using the table.
The Wei Li, Lei Zhang, David Zhang and Jingqi Yan, proposed the principal line based ICP Alignment
for Palmprint Verification [10]. First the modified finite Radon transform (MFRAT) is used to extract principal
line. The iterative closest point (ICP) alignment algorithm is used to compute the shifting, rotation and scaling
between the ROI images, and then presented an efficient way to combine the alignment result with the
competitive code for palmprint recognition.
The Cong Li, Fu Liu and Yongzhongzhang, proposed a method to extract the principal lines based on
their cartelistic of roof edges [11].In this at first gray adjustment and median filtering are used to enhance
contrast and weaken noise. Then, palm-lines are detected based on diversity and contrast. And an improved
Hilditch algorithm is used to do thinning, an edge tracking approach is applied to get rid of twigs and short lines,
and then, the broken lines are connected. Finally, the single pixel principal palm-line image is obtained.
The PatprapaTunkpien, SasipaPanduwadeethorn and SuphakantPhimoltares, proposed a simple and fast
method to extract the principle lines of palmprint by using consecutive filtering operations related to gradient
and morphological operators [12]. A gradient masks and closing operator are used to detect the lines. The results
are acceptable with 86.18 % accuracy.
The FengYue, WangmengZuo and David Zhang, proposed the iterative closest point (ICP) algorithm [13]
for palmprint alignment before matching. The palm-lines are extracted using steerable filter. However, due to
nonlinear deformation and inconsistency of extracted palm line feature, the ICP algorithm using only position
information would fail to obtain optimal alignment parameters. To improve its accuracy orientation feature is
used, which is more consistent than palm line, to make ICP registration more robust against noise.

II. The Proposed Framework

The proposed method consists of four steps
1. Image Acquisition
2. Image Preprocessing
3. Feature Extraction
4. Matching



Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 37 |

Fig.2 Proposed Architecture

2.1 Image Acquisition
The Poly U Palm print Database [14] is selected for experiment because it is widely used in palm print
biometric area.Poly U is the most widely used low-resolution palm print database for algorithmic research
considering recognition purposes. It is contained of 7752 images from 386 different persons. Persons provide
either the left or the right hand, but then not both. There are 20 samples for each in bitmap file format. Visually,
it is possible to identify more variability between images. The ROI of size 150x150 is cropped from each
original image to be initially used for extracting the principle lines with filtering procedures based on image
processing technique. Those image processing methods filtering algorithms that related to principle lines by
employing gradient masks and closing operator to detect the lines and the pre-processing including smoothing,
merging with binarization. It will display the principle lines and strong wrinkles.


Fig.3 Original imageFig.4 RegionofInterest (ROI)of palmprint

2.2Image Pre-Processing
2.2.1 Smoothing
To blur an image, smoothing filter is applied to eliminate small object in the image for noise reduction
purpose. It can also be used to get objects of interest which makes the image easy to detect. The smoothed
image can be implemented by applying the mask of standard size, 3x3, with the correlation function as shown in
Fig.6

1 1 1
1 1 1
1 1 1
Fig.5 3x3 smoothing filter mask
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 38 |

Fig.6Applying smoothing filter into an ROI

2.2.2 Edge detection
In our research, edge detection is implemented by gradient operator as the second filter in the whole
process. Firstly, masks of size 2x2 in two directions (0 and 90) are used as illustrated in Fig.7. With this small
size, edge can be easily detected with simply computation and less processing time. Subsequently, the smoothed
image is convolved with the masks in 0 and 90 respectively to enhance the edge in both directions. These
results are shown as Fig.7 and Fig.8 Finally, the magnitude of gradient is obtained by taking the root sum square
[15] of G
x
and G
y
:
Magnitude of gradient =
2
+
2

WhereG
x
andG
y
are image gradients in two directions.

G
x
=

G
y
=




Fig.7 First derivative mask in horizontal directions


Fig.8 First derivative mask in vertical directions
-2 -2
2 2
-2 2
-2 2
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 39 |

Fig.9 The magnitude gradient obtained from Fig.7 and Fig.8

2.2.3 Closing
Closing [16], the third operator, is the one of morphological operations which uses the basic operations
of dilation followed by erosion. The closing of image G by structuring element S is denoted by G S as follows
G S = (GS) S
Whereand are basic mathematical morphology techniques called dilation and erosion .The edge
image is performed by closing operator with disk-shape structuring element as demonstrated in Fig.10


Strel =


To smooth contours and fill small holes. After applying closing, the result of image is shown as Fig.11


Fig.11 After closing applying result image


Fig.12 Gradient of an image
0 1 0
1 1 1
0 1 0
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 40 |
3.2.4 Merging
After the image is detected, the principle lines and closing operation is used. Gradient image and
closing image are combined by merging with OR [17] operation as shown in Fig.13
This merged image obviously represents the contrast between object and background, so the
binarization technique with an automatically pre-defined threshold is employed for enhancing the object from
the background.


Fig.13 Binarization an image (ie) principle line and strong wrinkles will appear

2.3 Feature Extraction
Features play a significant role in image processing. The transformation of an image into set of features
known as Feature Extraction. Features, characteristics of the objects of interest, if selected carefully are
representative of the maximum relevant information that the image has to offer for a complete characterization
of a region. Feature extraction methodologies analyse objects and images to extract the most prominent features
that are representative of the various objects. Features are used as inputs to identify that assign them to the
person that they represent. In this work Gray Level Co-Occurrence Matrix (GLCM) features and Statistical
Property features are extracted from principal line extracted image.

2.3.1 GLCM (Gray Level Co-Occurrence Matrix) features
The GLCM features extracted in our research work are autocorrelation, contrast, correlation,
dissimilarity, homogeneity, entropy, maximum probability, sum of square, sum of variance, sum of entropy,
difference of entropy, and difference of variance. These features are developed by Haralick so it is also called
Haralick features .Some of the features described above can be calculated using the formula given below.

Energy
Energy returns the sum of squared elements in the Grey Level Co-Occurrence Matrix (GLCM). Energy is also
known as uniformity. The range of energy is [0 1]. Energy is 1 for a constant image. The formula for finding
energy is given in below equation:
= (, )
2
,
(1)
Contrast
Contrast returns a measure of the intensity contrast between a pixel and its neighbour over the whole
image. The range of Contrast is [0 (size (GLCM, 1)-1) ^2]. Contrast is 0 for a constant image. Contrast is
calculated by using the equation given below:
=
2
(, )
2
,
(2)
Correlation
Correlation returns a measure of how correlated a pixel is to its neighbour over the whole image. The
range of correlation is [-1 1]. Correlation is 1 or -1 for a perfectly positively or negatively correlated image.
Correlation is NaN (Not a Number) for a constant image. The below equation shows the calculation of
Correlation:
=

,
(3)
Where

and

are the means and standard deviations of P

i
and P
j
, the partial probability density
functions.


Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 41 |
Homogeneity
Homogeneity returns a value that measures the closeness of the distribution of elements in the GLCM
to the GLCM diagonal. The range of homogeneity is [0 1]. Homogeneity is 1 for a diagonal GLCM. The
homogeneity is evaluated using the equation:
=
(, )
1+
,
(4)

2.3.2 Statistical Properties of a palm image Features
The proposed work discusses about to measure properties of image regions. There are various
statistical measurements out of which part of our study and experiment are basic statistical properties of a palm
image and are Area, Bounding box and centric.
The description is for the measurement of a set of properties for each connected component (object) in
the binary image, BW. The image BW is a logical array; it can have any dimension
Area:
The Scalar can say an actual number of pixels in the region.
Bounding Box:
The smallest rectangle containing the region,
a 1-by-Q *2 vector, where Q is the number of image dimensions: ndims(L),
ndims(BW)
Centric:
It is 1-by-Q vector that specifies the center of mass of the region. Note that the first element of Centric
is the horizontal coordinate (or x-coordinate) of the center of mass, and the second element is the
vertical coordinate (or y-coordinate). All other elements of Centric are in order of dimension.
These basic statistical properties can be used to measure the statistical property of image region. The
calculated values of palm lines extracted image can be useful for palm matching technique. The matching can be
done by using basic statistical properties of palm and mostly useful on extracted palm lines image [18].

2.4 MATCHING
2.4.1 Euclidian Distance
One of the most popular similarity distance functions is the Euclidian distance. It is just the sum of the
squared distances of two vector values (xi,yi),
(5)
It is variant to both adding and multiplying all elements of a vector by a constant factor. It is also variant to the
dimensionality of the vectors, for example if missing values reduce the dimension of certain vectors produced
output will change. Given two data sets of features corresponding to the training and testing samples, a matching
algorithm ascertains the degree of similarity between them.

III. Experimental Results
The experiments were implemented in MATLAB 2012 software with Image Processing Toolbox and
on a machine with an Intel Core i3-2350M CPU@2.30GHz and 2.00 GB of RAM configured with
Microsoft Windows 7.The proposed model of this paper is tested on palmprint database collected by the Poly U
from 1000samples for100 different person left hand palm image (10 images for each person). Among them,
eight samples are used for training and remaining two samples are used for testing. For identification, each of
the palmprint images was matched with all of the other palmprint images in the database.
The performance of the proposed approach is evaluated using performance metrics i.e. FAR, FRR are
shown in Table I and comparative results of Accuracy and execution time using machine learning techniques are
shown in Table II.




(6)




(7)
FRR =
Number of rejected verification attempts for a qualified person

Number of all verification attempts for a qualified person
FAR =
Number of successful independent fraud attempts against a person

Number of all independent fraud attempts against a person
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 42 |
Table I. Comparative results of the FAR and FRR
P
e
r
s
o
n

Support vector
machine
Linear
Support vector
machine
RBF kernel
Extreme
Learning
Machine
Linear
Extreme
Learning
Machine
RBF kernel
FAR FRR FAR FRR FAR F
R
R
FAR F
R
R
1 1 1.98 1.5 1.97 0.96 5
2
0.02 9
9
2 0.5 1.9 1 1.98 0.97 5
2
0.05 9
9

Table II. Comparative results of the SVM and ELM classifiers

The comparative result shows that the predictive accuracy of Extreme Learning Machine linear is better
compared to Support Vector Machine with the kernels like linear and radial basis function. The execution time
taken by Extreme Learning Machine Linear is also less than the time taken by the different kernels of Support
Vector Machine.

IV. Conclusion
The proposed method to detect principle lines of palm print is with consecutive filtering operations.
Smoothing operation is used to remove image noise. Edge detector operation and closing operation are merged
to extract the principle lines. Binarization yields the binary principle line. The lines detected with the developed
scheme are used to extract textural information using Gray Level Co-occurrence Matrix and Statistical Property
Features. Euclidean distance is used for matching to identify the genuine person. The result shows that the
accuracy of Extreme Learning Machine linear is better compared to Support Vector Machine with the kernels
like linear and radial basis function. The execution time taken by Extreme Learning Machine Linear is also less
than the time taken by the different kernels of Support Vector Machine. In future, itwill improve the accuracy;
reduce the execution time for using other classification techniques.

REFERENCES
[1] D. Zhang,W.-K. Kong, J. You, and M.Wong, Online palmprint identification, IEEETrans.Pattern Anal. Mach.
Intell., vol. 25, no. 9, pp. 10411050, 2003
[2] A. Kong, D. Zhang, and M. Kamel, A survey of palmprint recognition, Pattern Recognition Letters, vol. 42,
pp. 14081418, 2009.
[3] P.H. Hennings, V. Kumar, and M. Savvides, Palmprint Classification Using MultipleAdvanced Correlation
Filters and Palm-Specific Segmentation, IEEE Proceedings of the Second International Conference on Knowledge
and Smart Technologies 2010 (July, 24-25, 2010)Transactions on Information Forensics and Security vol. 2. no3.
2007, pp. 613-622.
[4] M. Sakdanupab and N. Covavisaruch, An Efficient Approach for Automatic Palmprint Classification, Proceedings
of IEEE International Conference on Signal Image Technology and Internet Based Systems, 2008, pp. 229-234.
[5] D. S. Huang,W. Jia, and D. Zhang, Palmprint verification based on principal lines, Pattern Recognition, vol. 4,
pp. 13161328, 2008.
[6] K.Y.E. Wong, A. Chekima, J.A. Dargham, and G. Sainarayanan, Palmprint Identification Using Sobel Operator,
Proceedings of the 10th International Conference on Control, Automation, Robotics and Vision, Hanoi, Vietnam,
2008, pp. 1338-1341.
[7] L. Zhu and R. Xing, Hierarchical Palmprint Recognition Based on Major Line Feature and Dual Tree Complex
Wavelet Texture Feature, Proceedings of the Sixth international Conference on Fuzzy Systems and Knowledge
Discovery, 2009, pp. 15-19.
[8] Leqing Zhu, S.Z., RuiXing,Yin Zhang. Palmprint Recognition Based on PFI and Fuzzy Logic, in Fifth International
Conference on Fuzzy Systems and Knowledge Discovery,2008. FSKD '08. . 2008.
[9] Wei Jia, Yi-Hai Zhu, Ling-Feng Liu, De-Shuang Huang. Fast Palmprint Retrieval Using Principal Lines,
Proceedings of the 2009 IEEE International Conference on Systems, Man, and Cybernetics, San Antonio, TX,
Classifiers Kernels Execution Time
(seconds)

Accuracy (%)

SVM
Linear 1.772 88
Radial basis function(RBF) 1.825 98.1
ELM Linear 0.0082 99
Radial basis function(RBF) 0.0103 99
Palmprint Identification Based On Principle Line Using Machine Learning
| IJMER | ISSN: 22496645 | www.ijmer.com | Vol. 4 | Iss. 4 | Apr. 2014 | 43 |
USA - October 2009.
[10] W. Li, L. Zhang, D. Zhang, J.Q. Yan. Principal line based ICP alignment for palmprint verification, in: Proceedings
of the International Conference on Image Processing, 2009.
[11] Cong Li, Fu Liu and Yongzhongzhang. A Principal Palm-line Extraction Method for Palmprint Images Based on
Diversity and Contrast, Third International Congress on Image and Signal Processing (CISP2010),2010.
[12] PatprapaTunkpien, SasipaPanduwadeethorn, SuphakantPhimoltares. Compact Extraction of Principle Lines in
Palmprint using Consecutive Filtering Operations, Proceedings of the Second International Conference on
Knowledge and Smart Technologies ,July,24-25,2010.
[13] FengYue, WangmengZuo, David Zhang. ICP Registration Using Principal line and Orientation features for
Palmprint Alignment, Proceedings of 2010 IEEE 17th International Conference on Image Processing September
26-29, 2010, Hong Kong.
[14] PolyU Palmprint Database, http://www.comp.polyu.edu.hk /~biometrics.
[15] A. McAndrew, Introduction to Digital Image Processing with MATLAB, CourseTechnology, Boston,
Massachusetts, 2004.
[16] R.C. Gonzalez, R.E. Woods and S.L. Eddins.Digital Image Processing using MATLAB, Prentice-Hall, New
Jersey, 2002.
[17] MATLAB Student Center Tutorial, available athttp://www.mathworks.com/academia/student_
center/tutorials/intropage.html
[18] ShriramD.raut,Dr.vikasT.humbe.BiometricPalmprints features Matching for person Identification ,I.J modern
Education and computer science 2012,11,61-69.(http://www.mecs-press.org/)